You’re Wrong About DeepSeek
Oscar Gallo
Published on February 4, 2025
Lately, I’ve seen the ultimate meltdown over DeepSeek—a Chinese LLM hyped as a ChatGPT killer, supposedly “way better” than ChatGPT, and causing a ripple effect so massive it allegedly crashed the S&P 500 and Nvidia stocks.
So, let me cut through the noise: You’re wrong.
But first, let’s rewind a bit.
The Backstory
DeepSeek is a Chinese LLM that has achieved impressive results while using fewer resources and a lower-cost approach to training. It was developed by High-Flyer, a hedge fund co-founded by Liang Wenfeng in 2023.
The excitement? DeepSeek’s research shows groundbreaking optimizations, particularly in leveraging synthetic data (AI-generated instead of real-world data). This sparked hysteria over whether powerful GPUs are becoming obsolete—leading to Nvidia’s stock dip.
What Everyone Got Wrong
While DeepSeek’s methods are impressive, let’s set the record straight:
1️⃣ The GPU Narrative is Misleading
There’s no proof DeepSeek was trained on low-cost GPUs. In fact, Singapore—a country buying 22% of Nvidia’s high-end GPUs—is often used as a middleman for China to bypass sanctions. Which is more likely?
- 🔹 A 466-square-mile city-state, with no major AI startups, is reselling GPUs to China.
- 🔹 China magically trained DeepSeek on low-power GPUs.
2️⃣ DeepSeek Relied on ChatGPT
We know DeepSeek used ChatGPT-generated synthetic data for training. This drastically cuts costs, making DeepSeek’s progress dependent on OpenAI’s work. Without ChatGPT, DeepSeek wouldn’t be where it is today.
The Bottom Line
DeepSeek is a valuable addition to the AI landscape—competition benefits all of us, especially those building on top of these models. It’s also open-source, meaning you don’t have to use their servers to benefit from it (looking at you, “Open”AI).
Want to stay ahead in AI? Question the hype. Follow the data.